Skip to main content
Requirement factories for tool-use validation. Provides uses_tool, a Requirement factory that validates whether a model response includes a call to a specified tool — useful when you need to enforce tool invocation via rejection sampling rather than relying solely on the model’s tool_choice setting. Also provides tool_arg_validator, which validates the value of a specific argument to a named tool. Both accept either the tool’s string name or its callable.

​
Functions

​
FUNC uses_tool 

uses_tool(tool_name: str | Callable, check_only: bool = False) -> Requirement
Forces the model to call a given tool. Args:
  • tool_name: The tool that must be called; this can be either the name of the tool or the Callable for the tool.
  • check_only: Propagates to the Requirement.
Use tool_choice if the OpenAI tool_choice model option is supported by your model and inference engine. Returns:
  • A Requirement that validates whether the specified tool was called.

​
FUNC tool_arg_validator 

tool_arg_validator(description: str, tool_name: str | Callable | None, arg_name: str, validation_fn: Callable, check_only: bool = False) -> Requirement
A requirement that passes only if validation_fn returns a True value for the value of the arg_name argument to tool_name. If tool_name is not specified, then this requirement is enforced for every tool that was called. Args:
  • description: The Requirement description.
  • tool_name: The (optional) tool name to validate. When None, all tools are checked.
  • arg_name: The argument to check.
  • validation_fn: A validation function for validating the value of the arg_name argument.
  • check_only: propagates the check_only flag to the requirement.
Returns:
  • A Requirement that validates the specified tool argument.